Research On Key Technologies Of The Computation Offloading System In Mobile Cloud Environments

With the rapid development of computer and wireless network technologies,mobile devices(MDs)are becoming more and more popular and play a vital role in people's life and work.In recent years,benefiting from the improvements in chip manufacturing techniques,MD performance has been greatly improved.However,due to many limitations such as battery power,volume and heat dissipation,the MD itself has limited resources,which causes its development to encounter a resource bottleneck.To solve the resource limitation problem of MDs,people propose the concept of mobile cloud computing(MCC),and expect to offload computation tasks on MDs to the cloud to expand their available resources and to enhance their capability.MCC helps MDs break through their resource limitations,frees them from heavy computation tasks and makes them more responsible for connecting users and information domains.In MCC,users access the cloud through wireless networks,which seriously affect MCC performance.Compared with traditional cloud environments,mobile cloud environments are dynamic due to changes of wireless networks and user mobility.The dynamic mobile cloud environments bring challenges to the computation offloading system and put forward new requirements for its key technologies.This thesis studies key technologies of the computation offloading system in mobile cloud environments from the perspectives of users,cloud service operators,and cloud resource operators,and studies the offloading decision technology,the task admission control technology and the energy-efficient resource management technology.The main achievements of this thesis are as follows:(1)This thesis proposes a runtime offloading decision method based on memory-based immigrant adaptive genetic algorithm.With the help of the historical offloading strategies stored in memory,the decision method generates immigrant chromosomes to replace the original population's worst chromosomes during its execution,thereby enhancing its ability to adapt to environmental changes.The proposed method solves the multisite runtime offloading decision problem in mobile cloud environments.Simulation results show that,compared with other methods,the proposed offloading decision method effectively reduces the consumption of MDs and has obvious advantages in the scenario with large scale problems,weak running platforms or fast environmental changes.Moreover,for the problem of poor robustness and limited bandwidth in wireless networks,the usage of concurrent multiple transfer(CMT)in MCC is explored.Simulation results show that,compared with the traditional single-path data transmission,CMT can further reduce the consumption of MDs.(2)This thesis proposes a runtime offloading decision method based on machine learning(ML)and converts the offloading decision problem into the classification scheduling problem.Using the real-time characteristic of ML,the proposed method makes offloading decisions by the ML-based runtime scheduler.The decision method solves the runtime offloading decision problem in multi-user scenarios.An application partitioning algorithm based on adaptive simulated annealing genetic algorithm is proposed to generate training data for the ML-based scheduler efficiently.Simulation results show that the proposed method effectively reduces the consumption of MDs in multi-user offloading decision scenarios and its reduction is greater than the traditional methods.(3)This thesis establishes a task admission control model based on the long-term average criterion semi-Markov decision process,which considers the variation of wireless networks in mobile cloud environments.To develop an optimal admission control strategy,two strategy algorithms based on linear programming and reinforcement learning(RL)are proposed for different application scenarios.Simulation results show that the proposed strategy algorithms maximize the cloud service operator's profits while satisfying users' quality of service requirements,and the RL-based strategy algorithm can develop an approximate optimal strategy through the system simulation.(4)This thesis establishes a deterministic resource management model,which is a constrained optimization problem.To solve this problem,a deterministic strategy algorithm based on adaptive group genetic algorithm(AGGA)is proposed.On this basis,for the uncertain problem of wireless networks in resource management in mobile cloud environments,this thesis establishes a stochastic model that involves a stochastic optimization problem with chance constraints.To solve this problem,a stochastic strategy algorithm based on Monte Carlo simulation and AGGA is proposed.Simulation results show that the proposed deterministic strategy algorithm can develop an approximate optimal solution,and the proposed stochastic strategy algorithm can optimize the energy consumption while satisfying the chance constraints.